Search Results (1,708)

Machine learning models offer high discriminatory power for cross-sectional stroke-status classification, yet high-performing models often remain “black boxes” with little intrinsic interpretability. Bayesian networks (BNs) are probabilistic graphical models that offer intrinsic, auditable interpretability without the need for post hoc explanation. We specify a conditional Gaussian Bayesian network over eleven clinical variables, learned by constrained hill-climbing under epidemiologically derived arc constraints, parameterized by Bayesian estimation for discrete nodes and linear Gaussian regression for continuous ones, and queried for posterior stroke probability by Monte Carlo likelihood weighting. Using a 2 × 2 × 2 factorial design on 5109 patients, we crossed three preprocessing decisions—discretized versus continuous topology, median versus single imputation by chained equations (SICE), and no balancing versus synthetic oversampling (SMOTE)—yielding eight configurations, each assessed for discrimination, for calibration against a prevalence-only reference, and across 30 repeated stratified resamples of the entire pipeline analyzed by linear mixed model. Topology was the decisive choice: continuous nodes raised AUC by 0.050 (95% CI 0.038–0.063, p = 2.5 × 10⁻⁹), whereas imputation choice had no detectable effect (p = 0.741). Synthetic oversampling degraded discrimination and interacted antagonistically with topology, harming continuous networks five to six times more than discretized ones (p < 0.001). No configuration exceeded the prevalence-only reference in Brier skill score; a prior correction largely repaired the calibration intercept (−3.01 to −0.04) but not the slope or discrimination loss. The selected network (test AUC 0.811, 95% CI 0.761–0.853) showed no detected difference from logistic regression (AUC 0.819, p = 0.46) or XGBoost (AUC 0.822, p = 0.31). In conclusion, intrinsically interpretable Bayesian networks carried no measured accuracy penalty, though probability estimates require recalibration and longitudinal cohort validation. Full article

Urban expansion, peri-urban growth, and evolving mobility service systems are reshaping the functional links between urban and rural areas. As these relationships become increasingly complex, conventional urban–rural classifications may underestimate the spatial differences in everyday accessibility, environmental conditions, and demographic exposure. To capture these interrelated dimensions, this study develops a GeoAI-enabled Accessibility–Environment–Equity (AEE) framework to analyse the urban–rural continuum as a multidimensional functional space. Using Odense municipality as a case study, the framework incorporates network accessibility indicators, environmental benefit and pressure proxies, and demographic exposure indicators within a hexagonal tessellation. The indicators are combined into a standardised AEE feature matrix, interpreted through a principal component analysis, and classified using Gaussian mixture modelling with posterior-probability uncertainty mapping. The resulting typology shows a distinct but non-uniform core–periphery gradient: the central areas exhibit high accessibility and population density but lower green space availability and higher environmental pressure proxies, while the peripheral areas are greener but less accessible by non-car modes. The benchmarking against single-domain reference classifications reveals that the functional typology is primarily shaped by accessibility and demographic density structures rather than by greenness alone. This study advances urban–rural continuum (URC) research by demonstrating how GeoAI-enabled workflows can identify functional transition zones and accessibility–environment mismatches, supporting more evidence-based planning for medium-sized European cities and their surrounding urban–rural transition zones. Full article

Hand X-ray image segmentation is an important step in automated radiographic image analysis. However, conventional U-shaped segmentation networks often have relatively high model complexity, while variations in grayscale distributions across hand X-ray images may affect segmentation performance. To address these issues, this study proposes a lightweight hand X-ray image segmentation network, CSP-UNet (Cross-Stage Partial U-Net). The network integrates cross-stage partial feature processing into the U-Net encoder–decoder framework to reduce redundant feature computation and the number of model parameters while preserving effective feature representation. In addition, an adaptive Gaussian histogram-matching strategy is employed to reduce variations in grayscale distributions across X-ray images. CSP-UNet was evaluated on a dataset comprising 2000 hand X-ray images and compared with Classic U-Net, Res-UNet, Attention U-Net, and Swin U-Net. Experimental results show that CSP-UNet maintained comparable segmentation performance, achieving a Dice coefficient of 0.9927, PA of 0.9839, MPA of 0.9810, and mIoU of 0.9542, while requiring only 20.01 M parameters. Compared with Classic U-Net, CSP-UNet maintained a comparable Dice coefficient (0.9927 vs. 0.9910) while reducing the parameter count from 69.1 M to 20.01 M, corresponding to a reduction of approximately 71.04%. These results indicate that CSP-UNet maintains comparable segmentation performance while substantially reducing model complexity, offering a favorable trade-off between segmentation performance and model size for hand X-ray image segmentation. Full article

This study reanalyzes data from a previous investigation on critical thinking, metacognition, empathy, and psychological well-being by incorporating academic performance into a graphical-causal model. The aim was to compare two explanatory structures: a performance-mediated model, in which metacognition relates to psychological well-being through academic performance, and a direct metacognitive model, in which metacognition maintains a direct relation with well-being. The sample included 155 undergraduate psychology students. Critical thinking, metacognition, empathy, academic performance, and psychological well-being were modeled using directed acyclic graphs, conditional independence tests, Gaussian Bayesian networks, and bootstrap-based arc stability analysis. The initial performance-mediated DAG was not supported because the implied independence between metacognition and well-being conditional on performance was violated. The refined model, which included a direct metacognition–well-being path, was more compatible with the data. Metacognition showed the strongest estimated path to psychological well-being, whereas academic performance did not operate as the main mediating mechanism. Bootstrap results supported the stability of the metacognition–well-being adjacency. Overall, the findings suggest that critical thinking relates to psychological well-being mainly through metacognitive processes, under the assumptions of the specified DAG. Full article

Two paradigms dominate the literature on resilient strain behaviour of unbound granular materials (UGMs): empirical formulations, often lacking theoretical grounding, and machine learning (ML) models, operating as black boxes. This study proposes a hyperelastic strain energy potential—the KHP (Karasahin Hyperelastic Potential) model—deriving its volumetric component from the logarithmic compression relationship of critical-state soil mechanics and its shear component from a power-law distortional term. Analytical differentiation yields strains that inherently satisfy Maxwell’s symmetry, guaranteeing thermodynamic consistency. The model was calibrated to repeated-load triaxial data from sand-and-gravel and crushed limestone specimens and evaluated using Leave-One-Out Cross-Validation against three empirical models and two ML algorithms (Gaussian Process Regression, GPR, and a Neural Network). For axial strain, KHP ranked second only to GPR; for radial strain, it achieved the highest accuracy across both materials, outperforming all ML models. Its practical value was shown through a nonlinear finite element analysis of a flexible pavement section, reproducing the stress-dependent resilient behaviour of the base layer as a numerical demonstration of engineering usability, rather than a validation against measured field response. These findings show that a physical hyperelastic framework can match or exceed machine learning accuracy while preserving thermodynamic consistency, interpretability, and generalisability. Full article

Deep learning intrusion detectors for IoT networks are mostly evaluated on clean-data accuracy alone, leaving their behavior under perturbation unknown. This paper treats robustness as a tunable design parameter rather than an uncontrolled byproduct. The mechanism itself is not novel, since training with additive input noise penalizes the norm of the input gradient. What is new is its quantified use as an operational control, where one training-time parameter sets a measurable robustness radius. On ToN-IoT and Edge-IIoTset, we evaluate a lightweight CNN-BiLSTM-Attention detector of about 200 K parameters with mutual-information feature selection. Under test-time Gaussian noise and projected gradient descent on modifiable continuous features, clean-data accuracy does not guarantee stability after deployment. A strong random forest, a high-accuracy deep model and a parameter-matched transformer all collapse under small perturbations, so the fragility belongs to the detection task rather than to any model family. Noise-aware training establishes a controllable radius matched to the expected deployment perturbation, retaining about 88 percent accuracy where baselines collapse. It costs clean-data attack recall, which a validation-calibrated threshold restores without reducing robustness, and gradient-free attacks confirm the robustness is genuine. Robustness should therefore be reported alongside accuracy and efficiency as a standard axis for IoT intrusion detection. Full article

Shui Script is an ancient pictographic writing system with significant cultural and historical value. Automatic character detection is a fundamental step toward the digital preservation of Shui manuscripts, yet remains challenging due to slender strokes, densely distributed small characters, and complex background interference. To address these challenges, we propose DAC-Net, a lightweight dual-attention collaborative network based on YOLOv9. Specifically, the Dual-Attention Feature Enhancement Module (DAFEM) enhances weak stroke representations and preserves fine structural details, while the Decoupled Spatial-Channel Attention Fusion Module (DSAF) decouples spatial localization and channel semantic modeling to suppress background interference and improve feature discrimination. Extensive experiments demonstrate that DAC-Net achieves competitive performance against representative detection methods under multiple IoU thresholds while maintaining real-time inference efficiency. Notably, the performance improvement of DAC-Net over YOLOv9 is merely about 1% at IoU = 0.6 and IoU = 0.7. Robustness evaluations under brightness variation, motion blur, Gaussian noise, and partial occlusion further verify the effectiveness and generalization capability of the proposed method. Full article

Short-term multivariate forecasting of hydrological variables remains challenging because river systems exhibit nonlinear and time-dependent dynamics, complex relationships among water level, flow, and precipitation, and uncertainty arising from measurement errors and external disturbances. Although neural network models can learn nonlinear relationships among hydrological variables, their predictive performance often deteriorates in the presence of noise. Moreover, existing approaches rarely integrate the learning of long-term temporal dependencies and cross-variable relationships with a data assimilation mechanism capable of recursively updating state estimates and reducing forecast uncertainty. This limitation reveals the need for a robust forecasting framework that combines both capabilities. Therefore, this study aimed to develop and evaluate a hybrid Long Short-Term Memory–Ensemble Kalman Filter (LSTM–EnKF) model for short-term multivariate water-level forecasting under noisy conditions. The proposed framework extends a previously developed NARX–EnKF approach by replacing the NARX network with an LSTM architecture capable of learning nonlinear temporal patterns and relationships among water level, flow, and precipitation. The model was implemented using data from two hydrological stations located along the Atrato River in Colombia and configured to generate water-level forecasts with a two-day prediction horizon. The LSTM network generated the initial forecasts, whereas the EnKF assimilated the available observations to recursively update the estimated states and reduce forecast uncertainty. Model robustness was examined by introducing Gaussian white noise with variance levels of 0.001, 0.05, 0.10, and 0.20 to represent measurement uncertainty and external disturbances. Performance was evaluated using the root mean square error (RMSE), mean absolute error (MAE), and Nash–Sutcliffe efficiency (NSE). Across all evaluated noise levels, the LSTM–EnKF model outperformed the standalone LSTM model. Its total RMSE ranged from 0.1009 to 0.1929 m, compared with 0.1740 to 0.2356 m for the standalone LSTM, representing reductions of approximately 18.1–42%. The hybrid model achieved NSE values ranging from 0.9760 to 0.9992, whereas the standalone LSTM produced values between 0.9200 and 0.9776. Furthermore, the LSTM–EnKF reduced the MAE by approximately 51.9–56.2% across both outputs. These results indicate that integrating LSTM-based temporal learning with EnKF-based data assimilation improves short-term forecasting accuracy and robustness under noisy conditions. The developed framework provides a promising tool for supporting flood early-warning systems, flood-risk management, and the protection of riverine communities. Full article

The Anhui reach of the Huaihe River Basin features low-lying terrain, dense river networks, and high flood risk spatial heterogeneity, restricting regional socio-ecological sustainability. Traditional flood risk assessments suffer from incomplete indicators, subjective grading, and one-sided outputs when adopting single empirical or unsupervised clustering models. Based on disaster system theory, this study constructed a 14-indicator evaluation system covering hazard susceptibility, environmental sensitivity, and hazard-bearing vulnerability. Subjective-objective weights from Analytic Hierarchy Process (AHP) and entropy weight were fused via game theory equilibrium to build the Game Theory-AHP-Entropy Weight Coupled FCE Model (GAE-FCE Model) for comprehensive flood risk quantification. Meanwhile, principal component analysis (PCA) was utilized for indicator dimensionality reduction, and two unsupervised clustering models, namely, K-Means and Gaussian Mixture Model (GMM), were developed. Confusion matrix, Adjusted Rand Index, and correlation analysis were used to compare four model outputs. Individual models display weak correlation due to distinct risk characterization mechanisms; the ensemble scheme compensates single-model defects, achieving AUC = 0.9161, Recall = 0.9231, and F1 = 0.8889 under optimal weights, and a multi-model coupling framework was further proposed to obtain integrated flood risk zoning. Results show that five dominant factors (Distance to rivers, Topographic Roughness, NDVI, Distance to lakes, and Population Density) contribute over 59.6% of total weight. GAE-FCE reflects inundation characteristics yet lacks natural-factor comprehensiveness; clustering methods capture natural background differentiation but ignore socio-economic vulnerability, whereas ensemble coupling offsets individual model defects. High-risk zones are concentrated along major rivers, and low-risk areas mainly distribute in southern mountainous regions. This multi-model framework provides scientific support for flood prevention and territorial spatial optimization, advancing sustainable disaster risk reduction for the Huaihe River Basin and similar plain basins. Full article

This paper proposes a methodology for diagnosing the network environment in terms of disturbances and performance metrics in communication infrastructures. An approach for identifying GWN and PRN based on Artificial Intelligence is integrated. Naïve Bayes classification in Gaussian and Kernel probability density functions of datasets are created with confirmed verification using Resubstitution and Cross-Validation techniques. The high efficiency of the GRNNs created for GWN and PRN recognition has been established. An approach for finding an optimum in non-linear minimization with inequality constraints using Interior-point, SQP, Active-set and Genetic Algorithm techniques has been synthesized regarding an analytical model for SRT performance index prediction. Full article

APLNR is a therapeutically important G protein-coupled receptor (GPCR) implicated in cardiovascular and metabolic regulation; however, how agonist binding reorganizes receptor dynamics to promote signaling competence remains poorly understood. Here, we used the small-molecule agonist CMF-019 as a representative ligand and integrated Gaussian accelerated molecular dynamics (GaMD), Markov state models (MSMs), and neural relational inference (NRI) to characterize the conformational dynamics, kinetic organization, and allosteric communication of APLNR in apo and CMF-019-bound states. We found that CMF-019 binding altered structural flexibility in extracellular and intracellular regions while modifying collective motions within the transmembrane core, indicating a transition toward a more signaling-permissive dynamic state. MSM analyses further revealed that CMF-019 binding redistributed APLNR toward activation-related intermediate conformations and accelerated transitions among metastable states. Mechanistically, NRI uncovered extensive rewiring of the receptor communication network, in which CMF-019 binding strengthened transmembrane coupling and redirected signal propagation toward more convergent signaling routes linked to intracellular functional regions. Together, these findings suggest that CMF-019 promotes APLNR signaling competence through integrated kinetic and allosteric remodeling, revealing a dynamic mechanism by which an agonist can reorganize receptor communication prior to downstream coupling. Full article

Open-channel hydraulic analysis often requires repeated solution of implicit nonlinear equations and numerical integration of gradually varied flow (GVF), which can become computationally demanding in inverse, optimization, and sensitivity applications. This study develops a hydraulically informed artificial neural network (ANN) surrogate framework comprising ten independently trained models for normal and critical depths, alternative and conjugate depths, GVF-related water-surface behavior, and profile-based discharge inference. Hydraulic information is introduced through physically meaningful, and where appropriate dimensionless, variables and reference solutions derived from established governing equations or numerical hydraulic models, while ANN optimization remains data driven. Equation-generated test sets quantified surrogate fidelity, whereas HEC-RAS comparisons were treated as numerical hydraulic cross-verification rather than independent physical validation. The forward surrogates reproduced their reference mappings with high accuracy within the represented domains. Benchmarking against Random Forest, support vector regression, and Gaussian Process Regression for Models 1, 5, and 7 showed no universal algorithmic superiority; however, ANN provided a favorable trade-off among accuracy, relative-error robustness, compactness, and repeated-inference efficiency. For Model 7, ANN inference was approximately 249 times faster than conventional GVF calculation, with development cost recovered after about 1.03 × 10⁵ evaluations. Model 9 inferred discharge with a 4.75% error in the profile-based test. Observation-based assessment using 16 historical Missouri River stage–discharge events showed that direct HEC-RAS inversion yielded a MAPE of 44.71%, whereas observation-only ANN and hybrid HEC-RAS-ANN discrepancy correction reduced MAPE to 10.25% and 9.83%, respectively. The framework is therefore a computational complement to established hydraulic equations and numerical models, with broader field validation and explicit uncertainty treatment required for general deployment. Full article

Rapid and non-destructive identification of oilseed rape varieties and prediction of leaf Soil Plant Analysis Development (SPAD) value and water content are important for variety evaluation and plant-status monitoring. However, hyperspectral prediction is constrained by small sample size, costly reference measurements, and acquisition-batch differences. Samples were collected over four consecutive days. Day 1 samples were used for training and Wasserstein generative adversarial network with gradient penalty (WGAN-GP) generation. For Days 2–4, 20% of the samples from each day were combined to form a selection-validation set, while the remaining 80% were retained separately as Test sets 1–3. The improved WGAN-GP integrated principal component analysis–Gaussian mixture model (PCA–GMM)-based class assignment, a partial least squares regression (PLSR) consistency loss, physicochemical distribution control, and generated-sample selection. Under single-task modeling, the improved framework enhanced all three tasks across the test sets. On Test set 3, variety accuracy increased from 0.4672 to 0.6100, SPAD root mean square error of prediction (RMSEP) decreased from 6.2201 to 4.6303, and the water-content test-set correlation coefficient (r_p) increased from 0.6803 to 0.7842. The improved multi-task model also enhanced all three tasks. These findings support the framework within the investigated three-variety, four-day leaf setting. Full article

Reliability assessments of subsea systems are generally performed at two levels: structural reliability analysis of individual components and functional reliability analysis of the overall system using generic failure-rate databases. This study develops a component-to-system multi-scale framework that integrates these two levels for a subsea separation system operating at 3000 m water depth. At the component level, a Gaussian process regression (GPR) surrogate is developed from 474 finite element simulations of a vertical gravity separator. First-order reliability method (FORM) and Monte Carlo simulation (MCS) are then employed to assess the structural reliability, followed by a time-variant reliability analysis that accounts for corrosion effects. At the system level, the structural reliability model is integrated with functional failure rates through a Bayesian network that considers five equipment items and relevant risk-influencing factors. The surrogate model accurately predicts collapse pressure with an R² value of 0.996. The intact separator achieves a reliability index of 4.55, satisfying the DNV high-safety-class target, with the structural failure mode contributing only 0.0034% of the separator failure rate. Under a corrosion rate of 0.4 mm/year, the reliability index decreases to 3.12 over a 25-year service period. The structural failure rate crosses the DNV medium-safety-class target of 10⁻⁴ per year at year 12, increasing the structural contribution to the overall system failure frequency to 0.33%. Sensitivity analysis indicates that initial ovality and wall thickness are the most influential parameters affecting structural reliability and should therefore be prioritized in design and integrity management strategies. The framework is demonstrated on this physics-consistent dataset; validation against independent nonlinear finite element analyses and experimental collapse data is identified as the necessary next step before the results are used for design. Full article

We propose a data-driven procedure, based on convolutional variational autoencoders, to identify the presence of signal pulses in long time series. The dataset consists of synthetic waveforms, each composed of non-Gaussian noise and a log-normal-shaped signal of variable intensity, with a length of 10,000 samples. The model heavily compresses the input waveforms, allowing a direct study of such a reduced representation. After training for 150 epochs on 7500 waveforms, a region in the latent space where the network encodes time-series presenting only background noise emerges, allowing, in turn, to tag as candidates for containing a signal those falling outside. When applied to a test dataset of freshly generated waveforms, 100% of events with signal amplitudes well above the baseline noise are correctly labelled, and this fraction only decreases for amplitudes comparable with accidental noise pulses. This approach was designed to fully exploit the measurements in dual-phase Liquid Argon Time Projection Chambers, as the one of the Recoil Directionality experiment, built in the context of the Darkside project. Full article